People detection on video with darkflow

Simple script for people detection at video using darkflow library.

It takes a .mp4 video file and save it with bounding boxex around people.

To make it work you just need to start it and give a path to input and output files.

Examples:

python script.py input_video_path output_video_path

Net was trained on VOC data only with 16 epochs.

Accuracy may be improved with longer training.

Dependencies

Python3, tensorflow 1.0, numpy, opencv 3, darkflow

Installing

1. Download this repo.

2. Download weights (.ckpt) files from GoogleDisk and put them to ckpt folder.

Running the tests

To test script run

python script.py 'path_to_derictory'/test/TUD-Campus.mp4 'path_to_derictory'/test/TUD-Campus_result.mp4

Script must create file TUD-Campus_result.mp4 with bounding boxex around people.

• Test video from MOTChallenge

Built With

• Darkflow
• Tensorflow
• OpenCV